Documenting its behaviour better

[mkitti]

A good start is documenting its behaviour better - what is the buffer growth behaviour? How does one stop a transcoding? What is the "mark" in the buffer and how does one set and clear it?

I think documenting all supported behaviours is a good starting point for thinking about the API. For example, the way to stop a transcoding is to write a TOKEN_END to a stream. The obscurity of that suggests perhaps that should be a method instead